Download auth3in

WAREHOUSING WEB DATA
Jérôme Darmont, Omar Boussaid, and Fadila Bentayeb
Équipe BDD, Laboratoire ERIC
Université Lumière – Lyon 2
5 avenue Pierre Mendès-France
69676 Bron Cedex
France
E-mail: {jdarmont|boussaid|fbentaye}@univ-lyon2.fr
KEYWORDS
Web, Multimedia data, Integration, Modeling process, UML,
XML, Mapping, Data warehousing, Data analysis.
ABSTRACT
In a data warehousing process, mastering the data
preparation phase allows substantial gains in terms of time
and performance when performing multidimensional analysis
or using data mining algorithms. Furthermore, a data
warehouse can require external data. The web is a prevalent
data source in this context. In this paper, we propose a
modeling process for integrating diverse and heterogeneous
(so-called multiform) data into a unified format.
Furthermore, the very schema definition provides first-rate
metadata in our data warehousing context. At the conceptual
level, a complex object is represented in UML. Our logical
model is an XML schema that can be described with a DTD
or the XML-Schema language. Eventually, we have designed
a Java prototype that transforms our multiform input data
into XML documents representing our physical model. Then,
the XML documents we obtain are mapped into a relational
database we view as an ODS (Operational Data Storage),
whose content will have to be re-modeled in a
multidimensional way to allow its storage in a star schemabased warehouse and, later, its analysis.
1. INTRODUCTION
The end of the 20th Century has seen the rapid development
of new technologies such as web-based, communication, and
decision support technologies. Companies face new
economical challenges such as e-commerce or mobile
commerce, and are drastically changing their information
systems design and management methods. They are
developing various technologies to manage their data and
their knowledge. These technologies constitute what is called
“business intelligence”. These new means allow them to
improve their productivity and to support competitive
information monitoring. In this context, the web is now the
main farming source for companies, whose challenge is to
build web-based information and decision support systems.
Our work lies in this field. We present in this paper an
approach to build a Decision Support Database (DSDB)
whose main data source is the web.
The data warehousing and OLAP (On-Line Analytical
Processing) technologies are now considered mature in
management applications, especially when data are
numerical. With the development of the Internet, the
availability of various types of data (images, texts, sounds,
videos, data from databases…) has increased. These data,
which are extremely diverse in nature (we name them
“multiform data”), may be unstructured, structured, or even
already organized in databases. Their availability in large
quantities and their complexity render their structuring and
exploitation difficult. Nonetheless, the concepts of data
warehousing (Kimball 1996; Inmon 1996; Chaudhuri and
Dayal 1997) remain valid for multimedia data
(Thuraisingham 2001). In this context, the web may be
considered as a farming system providing input to a data
warehouse (Hackathorn 2000). Large data volumes and their
dating are other arguments in favor of this data webhouse
approach (Kimball and Mertz 2000).
Hence, data from the web can be stored into a DSDB such as
a data warehouse, in order to be explored by on-line analysis
or data mining techniques. However, these multiform data
must first be structured into a database, and then integrated in
the particular architecture of a data warehouse (fact tables,
dimension tables, data marts, data cubes). Yet, the classical
warehousing approach is not very adequate when dealing
with multiform data. The muldimensional modeling of these
data is tricky and may necessitate the introduction of new
concepts. Classical OLAP operators may indeed prove
inefficient or ill-adapted. Administering warehoused
multiform data also requires adequate refreshment strategies
when new data pop up, as well as specific physical
reorganization policies depending on data usage (to optimize
query performance). In order to address these issues, we
adopted a step-by-step strategy that helps us handling our
problem’s complexity.
In a first step, our approach consists in physically integrating
multiform data into a relational database playing the role of a
buffer ahead of the data warehouse. In a second step, we aim
at multidimensionally model these data to prepare them for
analysis. The final phase in our process consists in exploring
the data with OLAP or data mining techniques.
The aim of this paper is to address the issue of web data
integration into a database. This constitutes the first phase in
building a multiform data warehouse. We propose a
modeling process to achieve this goal. We first designed a
conceptual UML model for a complex object representing a
superclass of all the types of multiform data we consider
(Miniaoui et al. 2001). Note that our objective is not only to
store data, but also to truly prepare them for analysis, which
is more complex than a mere ETL (Extracting,
Transforming, and Loading) task. Then, we translated our
UML conceptual model into an XML schema definition that
represents our logical model. Eventually, this logical model
has been instantiated into a physical model that is an XML
document. The XML documents we obtain with the help of a
Java prototype are mapped into a (MySQL) relational
database with a PHP script. We consider this database as an
ODS (Operational Data Storage), which is a temporary data
repository that is typically used in an ETL process before the
data warehouse proper is constituted.
The remainder of this paper is organized as follows.
Section 2 presents our unified conceptual model for
multiform data. Section 3 outlines how this conceptual model
is translated into a logical, XML schema definition. Section 4
details how our input data are transformed into an XML
document representing our physical model. We finally
conclude the paper and discuss future research issues.
2. CONCEPTUAL MODEL
The data types we consider (text, multimedia documents,
represents a complex object generalizing all these data types.
Note that our goal here is to propose a general data structure:
the list of attributes for each class in this diagram is willingly
not exhaustive.
A complex object is characterized by its name and its source.
The date attribute introduces the notion of successive
versions and dating that is crucial in data warehouses. Each
complex object is composed of several subdocuments. Each
subdocument is identified by its name, its type, its size, and
its location (i.e., its physical address). The document type
(text, image, etc.) will be helpful later, when selecting an
appropriate analysis tool (text mining tools are different from
standard data mining tools, for instance). The language class
is important for text mining and information retrieval
purposes, since it characterizes both documents and
keywords.
Eventually, keywords represent a semantic representation of
a document. They are metadata describing the object to
integrate (medical image, press article...) or its content.
Keywords are essential in the indexing process that helps
guaranteeing good performances at data retrieval time. Note
that we consider only logical indexing here, and not physical
Figure 1: Multiform Data Conceptual Model
relational views from databases) for integration in a data
warehouse all bear characteristics that can be used for
indexing. The UML class diagram shown in Figure 1
issues arisen by very large amounts of data, which are still
quite open as far as we know. Keywords are typically
manually captured, but it would be very interesting to mine
them automatically with text mining (Tan 1999), image
mining (Zhang et al. 2001), or XML mining (Edmonds 2001)
techniques, for instance.
All the following classes are subclasses of the subdocument
class. They represent the basic data types and/or documents
we want to integrate. Text documents are subdivided into
plain texts and tagged texts (namely HTML, XML, or SGML
documents). Tagged text are further associated to a certain
number of links. Since a web page may point to external data
(other pages, images, multimedia data, files...), those links
help relating these data to their referring page.
Relational views are actually extractions from any type of
database (relational, object, object-relational — we suppose
a view can be extracted whatever the data model) that will be
materialized in the data warehouse. A relational view is a set
of attributes (columns, classically characterized by their
name and their domain) and a set of tuples (rows). At the
intersection of tuples and attributes is a data value. In our
model, these values appear as ordinal, but in practice they
can be texts or BLOBs containing multimedia data. The
query that helped building the view is also stored. Depending
on the context, all the data can be stored, only the query and
the intention (attribute definitions), or everything. For
On the other hand, if successive snapshots of an evolving
view are needed, data will have to be stored.
Images may bear two types of attributes: some that are
usually found in the image file header (format, compression
rate, size, resolution), and some that need to be extracted by
program, such as color or texture distributions.
Eventually, sounds and video clips are part of a same class
because they share continuous attributes that are absent from
the other (still) types of data we consider. As far as we know,
these types of data are not currently analyzed by mining
algorithms, but they do contain knowledge. This is why we
take them into account here (though in little detail),
anticipating advances in multimedia mining techniques
(Thuraisingham 2001).
3. LOGICAL MODEL
Our UML model can be directly translated into an XML
schema, whether it is expressed as a DTD or in the XMLSchema language. We considered using the XMI method
(Cover 2001) to assist us in the translation process, but given
the relative simplicity of our models, we proceeded directly.
The schema we obtained, expressed as a DTD, is shown in
<!ELEMENT COMPLEX_OBJECT (OBJ_NAME, DATE, SOURCE, SUBDOCUMENT+)>
<!ELEMENT OBJ_NAME (#PCDATA)>
<!ELEMENT DATE (#PCDATA)>
<!ELEMENT SOURCE (#PCDATA)>
<!ELEMENT SUBDOCUMENT (DOC_NAME, TYPE, SIZE, LOCATION, LANGUAGE?,
KEYWORD*, (TEXT | RELATIONAL_VIEW | IMAGE | CONTINUOUS))>
<!ELEMENT DOC_NAME (#PCDATA)>
<!ELEMENT TYPE (#PCDATA)>
<!ELEMENT SIZE (#PCDATA)>
<!ELEMENT LOCATION (#PCDATA)>
<!ELEMENT LANGUAGE (#PCDATA)>
<!ELEMENT KEYWORD (#PCDATA)>
<!ELEMENT TEXT (NB_CHAR, NB_LINES, (PLAIN_TEXT | TAGGED_TEXT))>
<!ELEMENT NB_CHAR (#PCDATA)>
<!ELEMENT NB_LINES (#PCDATA)>
<!ELEMENT PLAIN_TEXT (#PCDATA)>
<!ELEMENT TAGGED_TEXT (CONTENT, LINK*)>
<!ELEMENT CONTENT (#PCDATA)>
<!ELEMENT LINK (#PCDATA)>
<!ELEMENT RELATIONAL_VIEW (QUERY?, ATTRIBUTE+, TUPLE*)>
<!ELEMENT QUERY (#PCDATA)>
<!ELEMENT ATTRIBUTE (ATT_NAME, DOMAIN)>
<!ELEMENT ATT_NAME (#PCDATA)>
<!ELEMENT DOMAIN (#PCDATA)>
<!ELEMENT TUPLE (ATT_NAME_REF, VALUE)+>
<!ELEMENT ATT_NAME_REF (#PCDATA)>
<!ELEMENT VALUE (#PCDATA)>
<!ELEMENT IMAGE (COMPRESSION, FORMAT, RESOLUTION, LENGTH, WIDTH)>
<!ELEMENT FORMAT (#PCDATA)>
<!ELEMENT COMPRESSION (#PCDATA)>
<!ELEMENT LENGTH (#PCDATA)>
<!ELEMENT WIDTH (#PCDATA)>
<!ELEMENT RESOLUTION (#PCDATA)>
<!ELEMENT CONTINUOUS (DURATION, SPEED, (SOUND | VIDEO))>
<!ELEMENT DURATION (#PCDATA)>
<!ELEMENT SPEED (#PCDATA)>
<!ELEMENT SOUND (#PCDATA)>
<!ELEMENT VIDEO (#PCDATA)>
Figure 2: Logical Model (DTD)
instance, it might be inadequate to duplicate huge amounts of
data, especially if the data source is not regularly updated.
Figure 2.
We applied minor shortcuts not to overload this schema.
Since the language, keyword, link and value classes only
bear one attribute each, we mapped them to single XML
elements, rather than having them be composed of another,
single element. For instance, the language class became the
language element, but this element is not further composed of
the name element. Eventually, since the attribute and the
tuple elements share the same sub-element "attribute name",
we labeled it ATT_NAME in the attribute element and
ATT_NAME_REF (reference to an attribute name) in the tuple
element to avoid any confusion or processing problem.
4. PHYSICAL MODEL
We have developed a Java prototype capable of taking as
input a data source from the web, fitting it in our model, and
producing an XML document. The source code of this
application we baptized web2xml is available on-line:
http://bdd.univ-lyon2.fr/download/web2xml.zip.
We
view the XML documents we generate as the final physical
models in our process.
The first step of our multiform data integration approach
consists in extracting the attributes of the complex object that
has been selected by the user. A particular treatment is
applied depending on the subdocument class (image, sound,
etc.), since each subdocument class bears different attributes.
We used three ways to extract the actual data: (1) manual
capture by the user, through graphical interfaces; (2) use of
standard Java methods and packages; (3) use of ad-hoc
automatic extraction algorithms. Our objective is to
progressively reduce the number of manually-captured
attributes and to add new attributes that would be useful for
later analysis and that could be obtained with data mining
techniques.
The second step when producing our physical model consists
in generating an XML document. The algorithm’s principle
Image
User-prompted keywords:
– surf
– black and white
– wave
is to parse the schema introduced in Figure 2 recursively,
fetching the elements it describes, and to write them into the
output XML document, along with the associated values
extracted from the original data, on the fly. Missing values
are currently treated by inserting an empty element, but
strategies could be devised to solve this problem, either by
prompting the user or automatically.
At this point, our prototype is able to process all the data
classes we identified in Figure 1. Figure 3 illustrates how one
single document (namely, an image) is transformed using our
approach. A composite document (such as a web page
including pieces of text, XML data, data from a relational
database, and an audio file) would bear the same form. It
would just have several different subdocument elements
instead of one (namely plain and tagged text, relational view,
and continuous/sound subdocuments, in our example).
Eventually, in order to map XML documents into a
(MySQL) relational database, we designed a prototype
baptized xml2rdb. This PHP script is also available on-line:
http://bdd.univ-lyon2.fr/xml2rdb/. It operates in two
steps. First, a DTD parser exploits our logical model
(Figure 2) to build a relational schema, i.e., a set of tables in
which any valid XML document (regarding our DTD) can be
mapped. To achieve this goal, we mainly used the techniques
proposed by (Anderson et al. 2000; Kappel at al. 2000). Note
that our DTD parser is a generic tool: it can operate on any
DTD. It takes into account all the XML element types we
need, e.g., elements with +, *, or ? multiplicity, element lists,
selections, etc. The last and easiest step consists in loading a
valid XML document into the previously build relational
structure.
5. CONCLUSION AND PERSPECTIVES
We presented in this paper a modeling process for integrating
multiform data from the web into a Decision Support
XML Model
<?XML version=1.0?>
<!DOCTYPE MlfDt SYSTEM "mlfd.dtd">
<COMPLEX_OBJECT>
<OBJ_NAME>Sample image</OBJ_NAME>
<DATE>2002-06-15</DATE>
<SOURCE>Local</SOURCE>
<SUBDOCUMENT>
<DOC_NAME>Surf</DOC_NAME>
<TYPE>Image</TYPE>
<SIZE>4407</SIZE>
<LOCATION>gewis_surfer2.gif</LOCATION>
<KEYWORD>surf</KEYWORD>
<KEYWORD>black and white</KEYWORD>
<KEYWORD>wave</KEYWORD>
<IMAGE>
<FORMAT>Gif</FORMAT>
<COMPRESSION/>
<WIDTH>344</WIDTH>
<LENGTH>219</LENGTH>
<RESA>72dpi</RESA>
</IMAGE>
</SUBDOCUMENT>
</COMPLEX_OBJECT>
Figure 3: Sample Physical Model for an Image
Database such as a data warehouse. Our conceptual UML
model represents a complex object that generalizes the
different multiform data that can be found on the web and
that are interesting to integrate in a data warehouse as
external data sources. Our model allows the unification of
these different data into a single framework, for purposes of
storage and preparation for analysis. Data must indeed be
properly "formatted" before OLAP or data mining techniques
can apply to them.
Our UML conceptual model is then directly translated into
an XML schema (DTD or XML-Schema), which we view as
a logical model. The last step in our (classical) modeling
process is the production of a physical model in the form of
an XML document. XML is the format of choice for both
storing and describing the data. The schema indeed
represents the metadata. XML is also very interesting
because of its flexibility and extensibility, while allowing
straight mapping into a more conventional database if strong
structuring and retrieval efficiency are needed for analysis
purposes.
The aim of the first step in our approach was to structure
multiform data and integrate them in a database. At this
point, data are managed in a transactional way: it is the first
modeling step. Since the final objective of our approach is to
analyze multimedia data, and more generally multiform data,
it is necessary to add up a layer of multidimensional
modeling to allow an easy and efficient analysis.
One first improvement on our work could be the use of the
XML-Schema language instead of a DTD to describe our
logical model. We could indeed take advantage of XMLSchema’s greater richness, chiefly at the data type diversity
and (more importantly) inheritance levels. This would
facilitate the transition between the UML data representation
and the XML data representation.
Both the XML and XML-Schema formalisms could also help
us in the multidimensional modeling of multiform data,
which constitutes the second level of structuring. In
opposition to the classical approach, designing a
multidimensional model of multiform data is not easy at all.
A couple of studies deal with the multidimensional
representation and have demonstrated the feasibility of UML
snowflake diagrams (Jensen et al. 2001), but they remain
few.
We are currently working on the determination of facts in
terms of measures and dimensions with the help of imaging,
statistical, and data mining techniques. These techniques are
not only useful as analysis support tools, but also as
modeling support tools. Note that it is not easy to refer to
measures and dimensions when dealing with multimedia
documents, for instance, without having some previous
knowledge about their content. Extracting semantics from
multimedia documents is currently an open issue that is
addressed by numerous research studies. We work on these
semantic attributes to build multidimensional models.
The diversity of multiform data also requires adapted
operators. Indeed, how can we aggregate data that are not
quantitative, such as most multimedia attributes? Classical
OLAP operators are not adapted to this purpose. We
envisage to integrate some data mining tasks (e.g., clustering)
as new OLAP aggregation operators. Multiform data analysis
with data mining techniques can also be complemented with
OLAP’s data navigation operators.
Eventually, using data mining techniques help us addressing
certain aspects of data warehouse auto-administration. They
indeed allow to design refreshment strategies when new data
pop up and to physically reorganize the data depending on
their usage in order to optimize query performance. This
aspect is yet another axis in our work that is necessary in our
global approach.
REFERENCES
Anderson, R. et al. 2000. Professional XML Databases. Wrox
Press.
Chaudhuri, S. and Dayal, U. 1997. “Data Warehousing and OLAP
for Decision Support”. ACM SIGMOD International
Conference on Management of Data (SIGMOD 97), Tucson,
USA, 507-508.
Cover, R. 2001. “XML Metadata
http://xml.coverpages.org/xmi.html.
Interchange
(XMI)”.
Edmonds, A. 2001. “A General Background to Supervised
Learning in Combination with XML”, Technical paper,
Scientio Inc., http://www.metadatamining.com.
Hackathorn, R. 2000. Web farming for the data warehouse.
Morgan Kaufmann.
Inmon, W.H. 1996. Building the Data Warehouse. John Wiley &
Sons.
Jensen, M.R. et al. 2001. “Specifying OLAP Cubes On XML
Data”. 13th International Conference on Scientific and
Statistical Database Management (SSDBM 2001), Fairfax,
USA, 101-112.
Kappel, G. et al. 2000. “X-Ray – Towards Integrating XML and
Relational Database Systems”. 19th International Conference
on Conceptual Modeling, 339-353.
Kimball, R. and Mertz, R. 2000. The Data Webhouse: Building the
Web-enabled Data Warehouse. John Wiley & Sons.
Kimball, R. 1996. The data warehouse toolkit. John Wiley & Sons.
Miniaoui, S. et al. 2001. “Web data modeling for integration in data
warehouses”. First International Workshop on Multimedia
Data and Document Engineering (MDDE 01), Lyon, France,
88-97.
Tan, A.H. 1999. “Text Mining: The state of the art and the
challenges”. PAKDD 99 Workshop on Knowledge discovery
from Advanced Databases (KDAD 99), Beijing, China, 71-76.
Thuraisingham, B. 2001. Managing and Mining Multimedia
Databases, CRC Press.
Zhang J. et al. 2001. “An Information-Driven Framework for Image
Mining”, 12th International Conference on Database and
Expert Systems Applications (DEXA 2001), Munich, Germany;
LNCS 2113, 232-242.